enh: Extending api to support chirp_2 model, and support passing loca…

…tion (#1089)

.changeset/chatty-grapes-scream.md

@@ -0,0 +1,5 @@
+---
+"livekit-plugins-google": minor
+---
+
+Add support for google STT chirp_2 model.

livekit-plugins/livekit-plugins-google/livekit/plugins/google/models.py

@@ -3,7 +3,13 @@
 # Speech to Text v2
 
 SpeechModels = Literal[
-    "long", "short", "telephony", "medical_dictation", "medical_conversation", "chirp"
+    "long",
+    "short",
+    "telephony",
+    "medical_dictation",
+    "medical_conversation",
+    "chirp",
+    "chirp_2",
 ]
 
 SpeechLanguages = Literal[

livekit-plugins/livekit-plugins-google/livekit/plugins/google/stt.py

@@ -28,6 +28,7 @@
     utils,
 )
 
+from google.api_core.client_options import ClientOptions
 from google.api_core.exceptions import Aborted, DeadlineExceeded, GoogleAPICallError
 from google.auth import default as gauth_default
 from google.auth.exceptions import DefaultCredentialsError
@@ -81,6 +82,7 @@ def __init__(
         punctuate: bool = True,
         spoken_punctuation: bool = True,
         model: SpeechModels = "long",
+        location: str = "global",
         credentials_info: dict | None = None,
         credentials_file: str | None = None,
         keywords: List[tuple[str, float]] | None = None,
@@ -97,6 +99,7 @@ def __init__(
         )
 
         self._client: SpeechAsyncClient | None = None
+        self._location = location
         self._credentials_info = credentials_info
         self._credentials_file = credentials_file
 
@@ -132,9 +135,16 @@ def _ensure_client(self) -> SpeechAsyncClient:
             self._client = SpeechAsyncClient.from_service_account_file(
                 self._credentials_file
             )
-        else:
+        elif self._location == "global":
             self._client = SpeechAsyncClient()
-
+        else:
+            # Add support for passing a specific location that matches recognizer
+            # see: https://cloud.google.com/speech-to-text/v2/docs/speech-to-text-supported-languages
+            self._client = SpeechAsyncClient(
+                client_options=ClientOptions(
+                    api_endpoint=f"{self._location}-speech.googleapis.com"
+                )
+            )
         assert self._client is not None
         return self._client
 
@@ -150,7 +160,7 @@ def _recognizer(self) -> str:
             from google.auth import default as ga_default
 
             _, project_id = ga_default()
-        return f"projects/{project_id}/locations/global/recognizers/_"
+        return f"projects/{project_id}/locations/{self._location}/recognizers/_"
 
     def _sanitize_options(self, *, language: str | None = None) -> STTOptions:
         config = dataclasses.replace(self._config)

tests/pytest.ini

@@ -1,4 +1,5 @@
 [pytest]
 asyncio_mode = auto
+asyncio_default_fixture_loop_scope = "function"
 log_cli = true
 log_cli_level = DEBUG

tests/test_stt.py

@@ -38,6 +38,12 @@ def read_mp3_file(filename: str) -> rtc.AudioFrame:
 RECOGNIZE_STT = [
     deepgram.STT(),
     google.STT(),
+    google.STT(
+        languages=["en-AU"],
+        model="chirp_2",
+        spoken_punctuation=False,
+        location="us-central1",
+    ),
     openai.STT(),
     fal.WizperSTT(),
 ]
@@ -63,6 +69,12 @@ async def test_recognize(stt: agents.stt.STT):
     assemblyai.STT(),
     deepgram.STT(),
     google.STT(),
+    google.STT(
+        languages=["en-AU"],
+        model="chirp_2",
+        spoken_punctuation=False,
+        location="us-central1",
+    ),
     agents.stt.StreamAdapter(stt=openai.STT(), vad=STREAM_VAD),
     azure.STT(),
 ]